A5102838140- Natural Language Processing Techniques
- Topic Modeling
- Digital Media Forensic Detection
- Image Enhancement Techniques
- Advanced Steganography and Watermarking Techniques
- Statistical Methods in Epidemiology
- Image and Signal Denoising Methods
- Imbalanced Data Classification Techniques
- Multimodal Machine Learning Applications
- Text Readability and Simplification
- Video Surveillance and Tracking Methods
- Neural Networks and Applications
- Anomaly Detection Techniques and Applications
- Human Pose and Action Recognition
- Advanced Neural Network Applications
- Smart Agriculture and AI
- Generative Adversarial Networks and Image Synthesis
- Medical Coding and Health Information
- Time Series Analysis and Forecasting
- Image Processing Techniques and Applications
- Data Mining Algorithms and Applications
- Law in Society and Culture
- Machine Learning in Healthcare
- Network Security and Intrusion Detection
- Direction-of-Arrival Estimation Techniques
Ludong University
2024
Space Engineering University
2023
Alibaba Group (China)
2023
Jilin University
2015-2022
Qingdao Agricultural University
2022
Northeastern University
2019-2021
Centre de Recerca Matemàtica
2021
Computer Vision Center
2021
Shanxi University
2018-2021
Jilin Medical University
2014-2021
Traditional plant disease diagnosis methods are mostly based on expert diagnosis, which easily leads to the backwardness of crop control and field management. In this paper, improve speed accuracy classification, a detection classification method optimized lightweight YOLOv5 model is proposed. We propose an IASM mechanism efficiency model, achieve weight reduction through Ghostnet WBF structure, combine BiFPN fast normalization fusion for weighted feature up learning each layer. To verify...
With the development of location-acquisition technologies, there are a huge number mobile trajectories generated and accumulated in variety domains. However, due to constraints device environment, many recorded at low sampling rate, which increases uncertainty between two consecutive sampled points trajectories. Our task is recover high-sampled trajectory based on irregular low-sampled free space, i.e., without road network information. There major problems with traditional solutions. First,...
Bei Li, Yinqiao Chen Xu, Ye Lin, Jiqiang Liu, Hui Ziyang Wang, Yuhao Zhang, Nuo Zeyang Kai Feng, Hexuan Chen, Tengbo Yanyang Qiang Tong Xiao, Jingbo Zhu. Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1). 2019.
This paper presents results from the second Thermal Image Super-Resolution (TISR) challenge organized in framework of Perception Beyond Visible Spectrum (PBVS) 2021 workshop. For this edition, same thermal image dataset considered during first has been used; only mid-resolution (MR) and high-resolution (HR) sets have considered. The consists 951 training images 50 testing for each resolution. A set 20 resolution is kept aside evaluation. two evaluation methodologies proposed are also...
Large amounts of data has made neural machine translation (NMT) a big success in recent years. But it is still challenge if we train these models on small-scale corpora. In this case, the way using appears to be more important. Here, investigate effective use training for low-resource NMT. particular, propose dynamic curriculum learning (DCL) method reorder samples training. Unlike previous work, do not static scoring function reordering. Instead, order dynamically determined two ways - loss...
The receiver operating characteristic (ROC) curve is a representation of the statistical information from binary classification problems and key concept in machine learning data science. We study properties ROC curves their implications. relation between fundamental "fine" are clarified. Our discussions based on randomization method. theoretical discussion demonstrated using large dataset pregnancy outcomes doctor diagnoses pre-pregnancy checkups provided by Chinese Ministry Health vehicle...
The deep network model, with the majority built on neural networks, has been proved to be a powerful framework represent complex data for high performance machine learning. In recent years, more and studies turn nonneural approaches build diverse structures, Deep Stacking Network (DSN) model is one of such that uses stacked easy-to-learn blocks parameter-training-parallelizable network. this paper, we propose novel SVM-based (SVM-DSN), which DSN architecture organize linear SVM classifiers A...
Adverse pregnancy outcomes can bring enormous losses to both families and the society. Thus, outcome prediction stays a crucial research topic as it may help reducing birth defect improving quality of population. However, recent advances in adverse detection are driven by data collected after mothers having been pregnant. In this situation, if bad is diagnosed, parents will suffer physically emotionally. paper, we develop deep learning algorithm which able detect classify before getting We...
With the development of Internet applications, educational websites have been well developed. This means that more and students acquire knowledge through virtual classroom. The system will produce a large number students' learning behavior data. It doesn't make sense if you just take student data as pure We extract characteristics from these behavioral data, then use mining algorithm to train classifier. Then it predicts final grades students. Warning who risk failing exam in time according...
The majority of the classical dimensionality reduction methods can be unified into a graph-embedding-based framework. A fixed graph constructed in high-dimensional space has been extensively employed methods. However, often cannot characterize structure data owing to curse dimensionality. To solve this problem, we combine construction and coherent Thus, updated dynamically reduction. In existing based on framework, graphs are usually by type neighborhood relationship single clustering...
DeepFake digital images have serious negative impacts on news integrity, legal forensics, and social security. In order to detect the more accurately, a method based face recognition is proposed. Face image feature vectors are extracted by Facenet, Euclidean distances among of different calculated as classification principle. Then, machine learning algorithms trained perform binary real fake images. The experimental results Celeb-DF data set show that proposed has better detection effect...
In recent years, with the continuous upgrading of computer hardware and development deep learning technology, new multimedia tampering tools can make it easier for people to tamper faces in videos. Tampered videos produced by these may hardly be detected human, so we need effective method detect face-tampered Current popular video face technologies mainly include Deepfake technology based on self-encoder Face2face graphics. this paper, propose a detection full faces. Facenet algorithm is...
Scilab is the famous open source software. has been extensively received and it designed a system. So scilab good platform to develop some useful toolbox. Random Forests an excellent machine learning algorithm. It efficient process large data can solve unbalanced classification problems. used widely. While because of disadvantages, current toolboxes are not popular. In this paper, we toolbox with scilab. We test its performance apply handwritten numeral recognition. The objective work...
Telemedicine has launched among many countries around the world. This paper presents a framework of telemedicine diagnosis decision. Here our frame is based on Multi-Agent system and Bayesian network. The composed several groups agents providing flexible, applicable analysis environment. Diagnosis decision-making built network model basing probability physiological indices disease.
With the development of Virtual Reality technology and next Human-Machine Interaction technology, this paper focus on object motion detection skin color analysis, provide one kind hand gesture segmentation method based camera. This capture image from single camera to detect moving by time difference Gaussian module method, tracking region real time, then segment using specified features after is extracted. Using both, do static recognition template match extracting contour.This experiment...
In order to effectively detect whether digital images are spliced, a blind forensics method of image splicing based on deep learning is proposed. The uses high-pass filter preprocess the image, weakens negative influence content tampering forensic analysis, implements feature selection and classification convolutional neural networks(CNNs) realize real spliced images. Experiments Columbia detection evaluation dataset comparison with traditional methods show that proposed can achieve better accuracy.
Imaging device recognition is an important research hotspot in image tampering analysis. In recent years, it has received extensive and rapid development. Image analysis based on imaging devices field tampering, the of also become important. order to promote equipment, this paper summarizes discusses current main methods representative work equipment identification. This article compares similarities differences traditional related deep learning methods, respectively, details The principles...
Pedestrian detection based on images is one key technology of intelligent vehicles, and it also widely applied in robots, surveillance. This paper mainly focuses implementing a pedestrian system, which classified by linear SVM with optimized Hog (Histograms Oriented Gradients) as the extracted features. Then some experiments were done to find out that how changing resolution training set, times bootstrapping iterations different size steps sliding windows affect overall performance detecting systems.
In distance-selected imaging, the contrast of laser images is reduced due to long imaging distances, insufficient power, and atmospheric turbulence. An enhancement algorithm based on EnlightenGAN network proposed improve images. Firstly, are acquired using a distance selection pass system establish image dataset expand dataset, traditional used enhance mapping relationship between low-quality high-quality The global discriminator PatchGAN with improved VGG model regularize self-feature...
The hashtag is an effective tool to manage and distribute social media content in recent years. Most existing tag recommendation methods rely on user profiles improve the F1 score by roughly 20%. In final results, multimodal information accounts for 58% of total, while 42%. However, these neither provide a sufficient fusion method across modalities nor ignore visual from main sources tags. this paper, we propose novel model entitled Transformer Network with Object Detection (TNOD), which...